Windows 电脑安全清理:把磁盘清理做成一条可审计的流水线

Windows 电脑安全清理是 SkillHub 上的 community 类 Skill,版本 1.0.2,分类 it-ops-security,由 user_b2e0f551 发布,标签 requires_api_key=false。它把 Windows 磁盘清理拆成可信空间量测、只读审计、分类计划、一次性令牌授权、分阶段执行、终态核验六个阶段,最终输出 HTML 报告。内置红线白名单与统一删除网关,永久删除模式被永久禁用,全程本地执行,不联网、不上传数据。触发词包括清理电脑、C盘满了、卸载预装、开机自启、临时文件、释放磁盘空间。当前评分 100000,下载 1196,安装 0。页面未提供安装命令与配置字段,适合需要可留档清理记录、数据不出本机的场景,不适合需要彻底销毁数据的场景。

Read More
agent调用cli:把昂贵模型的执行环节外包给本地 CLI 代理

这是一个用于编排外部本地 CLI 代理的 Agent Skill。它解决的问题是:高成本模型已经设计好计划或框架之后,仍需由同一模型逐字填充内容、批量生成文件,前一步需要判断力,后一步只需要执行力,按同一 token 价格计费并不划算。该 Skill 负责把执行侧交给本地已装好的外部 CLI 代理,如 opencode、claude/claudecode、qwen、gemini、codex 或其他模型 CLI,并在委派过程中提供并发控制、结果校验、运行报告、模型披露和 CLI 安装检查。官方页面标注坐标 @user_b51927e1/cli-agent-orchestrator,版本 0.1.0,来源 community,分类 ai-agent,标签 requires_api_key=false,页面评分 100000、下载 523、安装 0。它只负责委派与管理,不负责方案设计本身,规划仍由主模型完成。

Read More
dsh-orca-agents: Scheduling Multiple Code Models via Local Orca

dsh-orca-agents is a DSH workflow plugin that leverages local Orca as the execution runtime, enabling the DSH Agent to act as a scheduler and support directly dispatching tasks such as those for Grok, Codex, and Claude in local environments. By automatically creating isolated worktrees and empty projects, the plugin enables isolated task execution and automatic wake-up without transferring code or screenshots remotely. Users only need to describe their requirements in the conversation; the plugin automatically creates repositories, invokes the Orca Agent, and notifies DSH after task completion. This plugin depends on the Orca desktop environment and a specific version, making it suitable for scenarios requiring localized and isolated code execution.

Read More
dsh-lark: Feishu Gateway Plugin for DeepSeek Harness

`dsh-lark` is a plugin for DeepSeek Harness (DSH) that enables interaction between a local agent and Lark (Feishu) via a long-lived WebSocket connection. The plugin does not require a public callback URL and supports multimodal input, file transfer, single-card status display, and tool approval, aiming to provide a “conversation-driven” development experience. By default, it restricts access using an `open_id` allowlist and includes commands such as `/steer` and `/approve`, making it suitable for running local agents and performing remote debugging or automated execution.

Read More